Reproducing JPM’s US Rates Strategy RV Trade Idea¶

Medium Post: https://medium.com/@cleeclee/reproducing-jpms-us-rates-strategy-rv-trade-idea-6b38e89791d0

Jay Barry (J.P. Morgan’s Co-head of U.S. Rates Strategy) and team recently initiated a 3.125% Nov-41/ 2% Nov-41 flattener in their flagship “U.S. Fixed Income Markets Weekly” report from 20 September 2024. Here’s the snippet:

Turning to relative value, we’ve noticed that originally issued 20-year bonds maturing in late-2041 and early-2042 have cheapened relative to our par curve, whereas originally issued 30-year bonds with similar maturities have outperformed. Clearly, the recent steepening of the Treasury curve supports wider yield spreads between higher-coupon OI 30s, and lower-coupon OI 20s with the same maturity, given significant duration differences. However, this dynamic is not explained by duration profiles: Figure 20 displays the securities in the 2040–2043 basket, sorted by modified duration and shows that late-41/ early-42 OI 30s appear rich relative to other securities with similar duration. In particular, the 3.125% Nov-41/ 2% Nov-41 curve appears 5.0bp too steep relative to the shape of the 15s/20s Treasury par curve (Figure 21). Hence, we recommend initiating 91:100 weighted 3.125% Nov-41/ 2% Nov-41 flatteners (see Trade recommendations).
In [27]:
import sys
sys.path.append("../")
In [28]:
from CurveInterpolator import GeneralCurveInterpolator
from CurveDataFetcher import CurveDataFetcher
from utils.viz import plot_usts, run_basic_linear_regression_df, plot_residuals_timeseries
In [29]:
import pandas as pd
from datetime import datetime

import matplotlib.pyplot as plt
import matplotlib.pylab as pylab
params = {
    "axes.titlesize": "x-large",
    "legend.fontsize": "x-large",
    "axes.labelsize": "x-large",
    "xtick.labelsize": "x-large",
    "ytick.labelsize": "x-large",
}
pylab.rcParams.update(params)

import seaborn as sns
sns.set(style="whitegrid", palette="dark")

import nest_asyncio
nest_asyncio.apply()

import warnings
warnings.filterwarnings('ignore', category=FutureWarning)
warnings.filterwarnings('ignore', category=pd.errors.SettingWithCopyWarning)

import plotly
plotly.offline.init_notebook_mode()

%load_ext autoreload
%autoreload 2

The autoreload extension is already loaded. To reload it, use:
  %reload_ext autoreload
In [30]:
curve_data_fetcher = CurveDataFetcher(use_ust_issue_date=True)
In [31]:
quote_type = "eod"
as_of_date = datetime(2024, 10, 3)

curve_set_df = curve_data_fetcher.build_curve_set(
    as_of_date=as_of_date,
    sorted=True,
    include_off_the_run_number=True,
    market_cols_to_return=[f"{quote_type}_price", f"{quote_type}_yield"],
    calc_free_float=True,
    use_github=True,
)

curve_set_df
Out[31]:
cusip security_type auction_date issue_date maturity_date time_to_maturity int_rate high_investment_rate is_on_the_run ust_label ... parValue percentOutstanding est_outstanding_amt corpus_cusip outstanding_amt portion_unstripped_amt portion_stripped_amt reconstituted_amt free_float rank
0 912797LS4 Bill 2024-09-05 2024-09-10 2024-10-08 0.013699 NaN 5.171 False 5.171% Oct-24 ... 7.725101e+08 0.003348 2.307550e+11 NaN 0.000000e+00 NaN 0.000000e+00 NaN -772.5101 16.0
1 912797LT2 Bill 2024-09-12 2024-09-17 2024-10-15 0.032877 NaN 5.053 False 5.053% Oct-24 ... 7.980483e+08 0.003458 2.307652e+11 NaN 0.000000e+00 NaN 0.000000e+00 NaN -798.0483 15.0
2 912797LU9 Bill 2024-09-19 2024-09-24 2024-10-22 0.052055 NaN 4.783 False 4.783% Oct-24 ... 7.665970e+08 0.003322 2.307462e+11 NaN 0.000000e+00 NaN 0.000000e+00 NaN -766.5970 14.0
3 912797LV7 Bill 2024-09-26 2024-10-01 2024-10-29 0.071233 NaN 4.783 True 4.783% Oct-24 ... 6.932014e+08 0.003005 2.306985e+11 NaN 0.000000e+00 NaN 0.000000e+00 NaN -693.2014 13.0
4 912797MA2 Bill 2024-09-05 2024-09-10 2024-11-05 0.090411 NaN 5.150 False 5.15% Nov-24 ... 4.049427e+08 0.002884 1.403951e+11 NaN 0.000000e+00 NaN 0.000000e+00 NaN -404.9427 12.0
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
390 912810TT5 Bond 2023-10-12 2023-10-16 2053-08-15 28.884932 4.125 NaN False 4.125% Aug-53 ... 8.606596e+09 0.120247 7.157430e+10 912803GU1 7.157430e+10 60954881.0 1.061942e+10 991450.0 52348.2848 4.0
391 912810TV0 Bond 2024-01-11 2024-01-16 2053-11-15 29.136986 4.750 NaN False 4.75% Nov-53 ... 4.567153e+08 0.006874 6.644364e+10 912803GW7 6.644364e+10 56812091.9 9.631553e+09 1679110.0 56355.3766 3.0
392 912810TX6 Bond 2024-04-11 2024-04-15 2054-02-15 29.389041 4.250 NaN False 4.25% Feb-54 ... 2.211754e+09 0.031064 7.119879e+10 912803GY3 7.119879e+10 58204025.7 1.299477e+10 792000.0 55992.2712 2.0
393 912810UA4 Bond 2024-07-11 2024-07-15 2054-05-15 29.632877 4.625 NaN False 4.625% May-54 ... 7.428279e+09 0.097202 7.642080e+10 912803HB2 7.642080e+10 68393441.8 8.027362e+09 1220467.0 60965.1628 1.0
394 912810UC0 Bond 2024-09-12 2024-09-16 2054-08-15 29.884932 4.250 NaN True 4.25% Aug-54 ... 4.964611e+09 0.095538 5.196489e+10 912803HD8 2.975507e+10 29061214.1 6.938548e+08 15000.0 24096.6028 0.0

395 rows × 25 columns

Build Par Curve Model¶

In [32]:
def liquidity_premium_curve_set_filter(curve_set_df: pd.DataFrame):
    
    # remove OTRs, olds, double olds, triple olds
    curve_set_filtered_df = curve_set_df[
        (curve_set_df["rank"] != 0)
        & (curve_set_df["rank"] != 1)
        & (curve_set_df["rank"] != 2)
        & (curve_set_df["rank"] != 3)
    ]

    # remove TBills
    curve_set_filtered_df = curve_set_filtered_df[
        curve_set_filtered_df["security_type"] != "Bill"
    ]

    # remove low free float bonds (< $5bn)
    curve_set_filtered_df = curve_set_filtered_df[
        curve_set_filtered_df["free_float"] > 5000
    ]

    # filter out bonds very close to maturity
    curve_set_filtered_df = curve_set_filtered_df[
        curve_set_filtered_df["time_to_maturity"] > 30 / 360
    ]

    # remove CTDs
    curve_set_filtered_df = curve_set_filtered_df[
        ~curve_set_filtered_df["cusip"].isin(
            [
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.625s 2026-09-15")["cusip"], # TU
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.125s 2027-09-30")["cusip"], # Z3N
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.25s 2029-02-28")["cusip"], # FV
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.25s 2031-06-30")["cusip"], # TY
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.375s 2034-05-15")["cusip"], # TN
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.625s 2040-02-15")["cusip"], # US
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.5s 2044-02-15")["cusip"], # TWE
                curve_data_fetcher.ust_data_fetcher.cme_ust_label_to_cusip("4.75s 2053-11-15")["cusip"], # UL
            ]
        )
    ]

    curve_set_filtered_df = curve_set_filtered_df.sort_values(by=["time_to_maturity"])
    
    return curve_set_filtered_df
In [33]:
# filter and fit bspline w/ knots are liquidity points
curve_set_filtered_df = liquidity_premium_curve_set_filter(curve_set_df=curve_set_df)

filtered_fitted_interpolator = GeneralCurveInterpolator(
    x=curve_set_filtered_df["time_to_maturity"].to_numpy(),
    y=curve_set_filtered_df[f"{quote_type}_yield"].to_numpy(),
)

fitted_bspline = filtered_fitted_interpolator.b_spline_with_knots_interpolation(
    knots=[0.5, 1, 2, 3, 4, 5, 7, 8, 9, 10, 11, 12, 15, 20, 25],
    k=3,
    return_func=True,
)
In [34]:
plot_usts(
    curve_set_df=curve_set_df,
    ttm_col="time_to_maturity",
    ytm_col=f"{quote_type}_yield",
    hover_data=[
        "issue_date",
        "maturity_date",
        "cusip",
        "original_security_term",
        "ust_label",
        f"{quote_type}_price",
        "free_float",
    ],
    ust_labels_highlighter=[("3.125% Nov-41", "red"), ("2.000% Nov-41", "blue")],
    zero_curves=[
        (
            fitted_bspline,
            "BSpline k=3 - Zero Filtered Fit",
        ),
    ],
    par_curves=[
        (
            fitted_bspline,
            "BSpline k=3 - Par FF",
        ),
    ],
    impl_spot_n_yr_fwd_curves=[
        (fitted_bspline, 1, "Impl Spots, 1y Fwd"),
    ],
    impl_par_n_yr_fwd_curves=[
        (fitted_bspline, 1, "Impl Par, 1y Fwd"),
    ],
    title=f"All USTs - using {f"{quote_type}_yield"} - as of {as_of_date}"
)

Fetching historical curve sets to regress 3.125% Nov-41/ 2% Nov-41 flattener vs our fitted model over time¶

  • Looking at a larger timeframe than what JPM looked at
  • From date is the first day of 2% Nov-41
In [35]:
start_date = datetime(2022, 1, 31)
end_date = datetime(2024, 10, 3)

curve_sets_dict_df, fitted_curves_dict = curve_data_fetcher.fetch_historical_curve_sets(
    start_date=start_date,
    end_date=end_date,
    fetch_soma_holdings=True,
    fetch_stripping_data=True,
    calc_free_float=True,
    fitted_curves=[("LPF", "eod_yield", liquidity_premium_curve_set_filter)],
)

FETCHING CURVE SETS...: 100%|██████████| 848/848 [00:14<00:00, 58.88it/s]
AGGREGATING CURVE SET DFs: 100%|██████████| 848/848 [00:23<00:00, 36.08it/s]

Better data structure to fetch specific CUSIP timeseries data¶

  • Comparing different curve building methods on different filtering strats
In [36]:
cusip_timeseries = {}

for dt in curve_sets_dict_df.keys():
    fitted_bspline = fitted_curves_dict[dt]["LPF"].b_spline_with_knots_interpolation(
        knots=[0.5, 1, 2, 3, 4, 5, 7, 8, 9, 10, 11, 12, 15, 20, 25],
        k=3,
        return_func=True,
    )
    curr_curve_set_df = curve_sets_dict_df[dt][["cusip", "time_to_maturity", "eod_yield"]]
    curr_curve_set_df["fitted_yield"] = fitted_bspline(curr_curve_set_df["time_to_maturity"])
    curr_curve_set_df["fitted_spread"] = curr_curve_set_df["eod_yield"] - curr_curve_set_df["fitted_yield"]
    curr_curve_set_df["fitted_15s20s"] = fitted_bspline(20) - fitted_bspline(15)
    for _, row in curr_curve_set_df.iterrows():
        if row["cusip"] not in cusip_timeseries:
            cusip_timeseries[row["cusip"]] = []

        cusip_timeseries[row["cusip"]].append(
            {
                "Date": dt,
                "eod_yield": row["eod_yield"],
                "fitted_yield": row["fitted_yield"],
                "fitted_spread": row["fitted_spread"],
                "fitted_15s20s": row["fitted_15s20s"],
            }
        )
In [37]:
label1 = "3.125% Nov-41"
cusip1 = curve_data_fetcher.ust_data_fetcher.ust_label_to_cusip(label1)["cusip"]

label2 = "2.0% Nov-41"
cusip2 = curve_data_fetcher.ust_data_fetcher.ust_label_to_cusip(label2)["cusip"]

cusip1_df = pd.DataFrame(cusip_timeseries[cusip1])
cusip2_df = pd.DataFrame(cusip_timeseries[cusip2])

spread_df = pd.DataFrame({
    "Date": cusip1_df["Date"],
    label1: cusip1_df["eod_yield"],
    label2: cusip2_df["eod_yield"], 
    f"{label1} / {label2}": cusip2_df["eod_yield"] - cusip1_df["eod_yield"],
    "fitted_15s20s": cusip1_df["fitted_15s20s"]
})
In [38]:
plt.figure(figsize=(20, 10))
plt.plot(spread_df["Date"], spread_df[label1], label=label1)
plt.plot(spread_df["Date"], spread_df[label2], label=label2)
plt.xlabel("Date")
plt.ylabel("Yield")
plt.title(f"{label1} vs {label2}")
plt.legend()
plt.grid(True)
plt.show()

plt.figure(figsize=(20, 10))
plt.plot(spread_df["Date"], spread_df[f"{label1} / {label2}"], label=f"{label1} / {label2}")
plt.xlabel("Date")
plt.ylabel("Yield (bps)")
plt.title(f"{label1} / {label2} Spread")
plt.legend()
plt.grid(True)
plt.show()

plt.figure(figsize=(20, 10))
plt.plot(spread_df["Date"], spread_df["fitted_15s20s"], label="Fitted 15s20s")
plt.xlabel("Date")
plt.ylabel("Yield (bps)")
plt.title("15s/20s Treasury par curve")
plt.legend()
plt.grid(True)
plt.show()
No description has been provided for this image
No description has been provided for this image
No description has been provided for this image
In [39]:
r = run_basic_linear_regression_df(
    df=spread_df,
    x_col="fitted_15s20s",
    y_col=f"{label1} / {label2}",
)
plot_residuals_timeseries(df=spread_df, results=r)
plot_residuals_timeseries(df=spread_df, results=r, plot_zscores=True)

                                 OLS Regression Results                                
=======================================================================================
Dep. Variable:     3.125% Nov-41 / 2.0% Nov-41   R-squared:                       0.126
Model:                                     OLS   Adj. R-squared:                  0.125
Method:                          Least Squares   F-statistic:                     95.77
Date:                         Fri, 04 Oct 2024   Prob (F-statistic):           3.22e-21
Time:                                 12:52:24   Log-Likelihood:                 1656.7
No. Observations:                          666   AIC:                            -3309.
Df Residuals:                              664   BIC:                            -3300.
Df Model:                                    1                                         
Covariance Type:                     nonrobust                                         
=================================================================================
                    coef    std err          t      P>|t|      [0.025      0.975]
---------------------------------------------------------------------------------
const             0.0216      0.003      8.026      0.000       0.016       0.027
fitted_15s20s     0.0896      0.009      9.786      0.000       0.072       0.108
==============================================================================
Omnibus:                       17.317   Durbin-Watson:                   0.043
Prob(Omnibus):                  0.000   Jarque-Bera (JB):               21.874
Skew:                           0.284   Prob(JB):                     1.78e-05
Kurtosis:                       3.682   Cond. No.                         12.7
==============================================================================

Notes:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.
No description has been provided for this image
No description has been provided for this image
No description has been provided for this image
In [ ]: